insufficient virtual memory

[avalaki]

I composed 4-meshed fire compartment with the alignment matches,
resulting in 1.8 million cells , but FDS running (version 5.1.4) shows
the following message;.

forrtl severe <41> insufficient virtual memory, stack trace terminated
abnomally.

I upgraded 4 GB RAM to Window XP professional SP2, Intel(R) and 2.21
GHz.
I also tried to increase the virtual memory up to " initial size of
3070" and " max. size of 4096" with the available size of c: 183,290MB

I really appreciate any advice and help to deal with this unexpected
enemy.

[Kevin]

You do not have enough RAM to run this case. I suggest you use four
computers and run in parallel with the MPI version of FDS. This is
precisely why we developed it.

[Glenn Forney]

Also note, with Windows XP your computer won't even see all 4GB of
memory - it will ONLY SEE 3GB . To see all 4GB you need to install
Windows x64.

--
Glenn Forney
National Institute of Standards and Technology
100 Bureau Drive, Stop 8663
Gaithersburg MD 20899-8663

Telephone: (301) 975 2313
FAX: (301) 975 4052

Pre-decisional and sensitive information. Not for attribution, distribution, or reproduction.

[avalaki]

Thanks Kevin,

I want to confirm your advice for the number of computers. Is it only
because I used 4-mesh input file regardless of 1.8 million cells?
Is there any recommended practice (experience) to the number of cells
for each mesh as well as total number of cells, even if there is a
remark at user guide that "remember, there is no limit to how big or
how long FDS calculation can be - it depends on the resources of the
computer"

Thanks again.

> > enemy.- Hide quoted text -
>
> - Show quoted text -

[dr_jfloyd]

If you have a case that will not run on a single computer due to
memory limitations, you have two options:

1) Reduce the number of grid cells
2) Split into multiple meshes and use multiple computers

Kevin suggested number 2 as you indicated that you had 4 meshes.

Multiple meshes does add some memory overhead for the arrays that
handle the boundary exchange info between meshes. This penalty is
proportionately worse for cases where the meshes have a high surface
area to volume ratio at the interface (i.e. a tall, thin mesh vs. a
cubical mesh).

[Apollo]

I have ran a couple of models using up to 2.3Mil cells (3 meshes) on a
2.4GHz Pentium Extreme with 4x 1G DDR ram. So the nos of cells I feel
is not the problem. Try to control the nos of output generation,
sampling timestep, etc which is taking bulk of memory.

[dr_jfloyd]

Output generation uses very little memory and memory usage is not
impacted at all by the sampling timestep.

[Apollo]

Thanks for the clarification. But like I've said, it's not an issue
for me. Perhaps because I'm using dedicated PC. There's only the OS
and FDS software inside. I've not installed other software.
Occassionally, I will clean the disk and defrag them.

Not sure if this is the case for the Thread Starter.

[Bak, Daniel]

Apollo,

I have recently experience the condition of insufficient virtual memory.
I also have 1 dedicated computer with 4 GB of memory.

Here what I have done and it did help a bit. I went into the windows
setup and maximized the virtual memory allocation. The second I have
done is reduce the number of particles by specifying PARTICLE_PART=10
instead of 1. The part5 file was reduced appreciably.

This is my 2 cents.

Daniel

[dragonit]

Well , I was wandering if the 1.8 mil cells is a total of the cells u
have in your input file
or a 1.8 million cells each?

In General:
When running a Win Xp 32bit it is true that the OS is not "seeing"
over 3giga of RAM
But, I have managed to run a simulation that utilized 6 giga RAM on a
winxp 32 bits
the problem with 32bit Os (a.k winXP 32bit..) is that each process
that is running can not exceed 2 Giga of RAM
But u can, by using the MPI ver run 4 meshes that each will hold a
memory up to 2 gigas (theoretically, in real life it's more like 1.5
Giga per process to leave air for the OS.)

In short :
each 1 million cells will require about 1 Giga of RAM,
each additional mesh will coast you in some memory overhead,
and the more reactions and activities their is to calculate inside the
mesh has an additional memory coast.

So if you can run 4 threads (4 meshes) using the mpi ver. And watch
your task manager to monitor the ram use, you can see if any one of
the fds_mpi processes are exceeding the 1.9 gigas of ram , that's the
point win xp will not tolerate .

If so, u should split your meshes to smaller ones

and well the more processors you have the faster it will run :)
so running this simulation on as many processors as the number of
meshes
(could also be a quad core ) will work better.

[Kevin]

How can you have 6 gigabytes of RAM on a 32 bit machine? Regardless of
the number of processors, there can only be 4 GB of addressable memory
for a 32 bit application. Does the computer have multiple banks of
RAM?

[Glenn Forney]

Click on the "My Computer" icon on your desktop with the right mouse
button. You should see the amount of Ram that Windows "thinks" you
have. On my computer, the amount shown is 3.25GB even though I actually
have 4GB installed.

--

[dragonit]

Yes that is correct ,
Windows can “see” only 3.25 Giga of ram
But In this system I use a 4x2giga slots on dual channel ram
and by using a Large memory swatch “/PAE” at the BOOT.ini
configuration I manged to stretch the ram limits a bit. But that was
just for fun :)

> Pre-decisional and sensitive information.   Not for attribution, distribution, or reproduction.- Hide quoted text -

[dr_jfloyd]

Various things can limit the amount of memory available to jobs. Some
computers shadow BIOS and other functions into RAM which reduces the
memory for the OS. Window XP itself will split RAM between the kernel
and codepage space for programs. So on a 4 GB machine, XP will
reserve 2 GB for the kernel leaving 2 GB for jobs. There is a boot
switch that can be set (/3GB) that will not let the kernel take more
than 1 GB (3 GB for jobs), but it sometimes leads to other problems.

> Pre-decisional and sensitive information.   Not for attribution, distribution, or reproduction.- Hide quoted text -